top of page

CERTH Visual Computing Lab solutions

CERTH Visual Computing Lab solutions

CERTH is participating in INFRASTRESS with the Visual Computing Lab and provides two components related to physical threats and hazards detection and one component on cyber and physical threat intelligence and prediction. 


The first is about “Relevant event and anomaly detection from video surveillance” where camera streams from the industrial plants and sites are analysed in real-time to provide alerts on anomalous events compared to the typical everyday situation, or predefined events that need special handling. The component is combining several deep learning architectures in one single software package that enables the monitoring of the situation for a set of issues. The detection of objects of critical importance such as misplaced luggage or the detection of a gun are also part of this component. 


The second component is also analyzing camera streams but with the focus on human activities and behaviors. The “People behavior analysis from visual data” component is identifying people in the camera streams and analyzing activities such a person holding a gun, or people running on panic, people fighting etc. to trigger the appropriate alerts. The re-identification of people seen from one camera to another is also a very useful tool to establish better understanding of the typical or anomalous movement of people in the monitored site, as well as trace back a suspicious person and identify possible locations that were sabotaged or affected.  Counting of people is also part of this component. 


Spatiotemporal forecasting of attacks and vulnerabilities is the cyber-physical threat intelligence component CERTH is providing. The aim of this component is to consume records and activities from multiple sensors and people reports in order to provide forecasts of the way a situation will progress both in temporal as well as in spatial manner. A map of the sensor recordings is required so that heatmaps of the forecasts can be are generated. Additional to the map of sensor recordings, a graph representation of different linked locations and movements of people are used to build a graph analysis layer that adds constraints on the way a phenomenon is evolving. 


Some example results using publicly available datasets are found in the following figures. 

  • Noir Icône YouTube
  • Noir Twitter Icon
  • Noir LinkedIn Icône
bottom of page